Scores increase since the server is scheduled for deletion

Server operators
1

Hello
As explained in my last post, I activated an NTP server on a GNSS correction service base from the Centipede network. This base is built around a GNSS receiver with a u-blox RTK ZED-F9P chip and a Raspberry Pi 3B+ computer. It is located on Réunion Island (Indian Ocean, Africa zone).

I registered the server with the NTP Pool on May 23, 2025, at 03:48 UTC : https://www.ntppool.org/scores/193.253.106.36

Unfortunately, I have observed two issues since then:

First, the GNSS correction service has been hanging regularly for short periods since the NTP server was registered:

Capture d’écran_2025-05-26_07-29-11
Capture d’écran_2025-05-26_07-29-111869×475 10.9 KB

Second, the score has consistently remained below 10.

I haven’t been able to identify the cause of these issues, but I suspect that either the bandwidth is too low, or the Raspberry Pi is not powerful enough to handle the workload.

So, I decided to schedule the server for deletion on May 28. This request was submitted on May 25 around 14:00 UTC. And, very strangely, since then:

  • The GNSS base has stopped hanging (see graph above).
  • The score has risen and stays since above 15 :

scores
scores1024×486 40.9 KB

How do you explain this behavior? Is it a kind of message saying “please don’t delete me”?
If I cancel the deletion, will these issues be back?

2

Looks like your server can’t handle the request if it’s in pool (S >= 10). The monitoring don’t get response from your server and the score drops below 10. The server left the pool. Monitoring can reach it again and the score rises.
Up and down.
If you set the server for deletion it will slowly dissappear from the pool list but the monitoring will do the job until it’s deleted.

You can cancle the deletion and set the server to monitoring only. I bet the score goes up to 20

3

Thanks for your answer.
I understand that when the server goes inside the pool, it receives connection attempts from NTP clients that it can’t handle.
How can I check what might be causing this? (Bandwidth issue, processing speed, or something else?)
I suppose there’s no point in leaving a server in monitoring-only mode?

4

I didn’t see any GNSS issues (e.g., bad time stamps). As apuls suggested, loss is a problem.
We don’t know whether the loss occurs in the server or in the transport network.

Does this server use an IP tunnel? Tunnels can be a problem: NTP requests from a client located in Africa may experience round-trip delay of 1/2 second or higher.

5

I don’t think so. This issue seems related to the one described here : Joining the pool kills my Internet
I had a similar issue this morning when I tried to reactivate the server, which nearly shut down internet access.

6

What kind of router do you have? It might be that the router tries to keep track of all the connections but gets overwhelmed by the number of requests. In this case the problem is not your Raspberry but your router. The solution is to not keep track of the connections. See iptables “notrack” and other options as discussed in the topic you linked.

7

Is the server realy located in africa ?
All IP Geolocation DB say it’s france not africa.

If your server is running in an underserved zone you can run into this trouble realy easy.

Did you tried the loweset NetSpeed and look what happen ?

8

Réunion is an island in the Indian Ocean which is (still) an overseas territory of France. So both the geolocation and the ntp zone are right :slight_smile:

9

It is a NETGEAR Router DGN1000v3.
I’ll see tomorrow if there is a notrack option.

10

The pool infrastructure uses MaxMind data, which points to Africa:

Screenshot_20250526_195346

In the fine print (aka web service interface), it says"is_in_european_union": true

@harlock, that question is still open, and relevant.

11

I briefly canceled deletion and put the connection speed to 1.5 mb here (arrow) :

hang
hang691×339 66.5 KB

But my internet connection hanged just after that and I need to put the server to monitor only to fix the situation. So may be this is a router issue after all.

12

If you set it for deletion in a couple of days (ASAP, in other words) I think it’s effectively removed from the pool immediately. I recall seeing mention the server is removed from the pool two weeks before the scheduled deletion, but perhaps that’s the beginning of a fading presence, I do not know.

13

The lowest setting is 512Kbit. Please try that.

But in an underserved zone, that might still be too much, though.

Someone once proposed to add even lower netspeed settings to choose from, at least for known underserved zones, but that proposal only got an eye roll emoji…

But being the only server in your zone, the netspeed setting probably doesn’t matter at all. As there is no one else to share the load with, your server always gets 100%.

(There are ways to get lower settings even today, but that requires some “low-level” access to the internals of the management portal.)

14

Yeah i know - But it’s a local copy and not an online lookup. Therefore the DB is sometimes old / inaccurate.
I’m using for lookups iplocation.net

15

I found that offline copy to be updated rather frequently. Not sure whether it follows the twice-weekly release by MaxMind. But a correction of mine appeared in the pool infrastructure just a few days after it made it into the online database (and corresponding offline database edition downloadable from MaxMind).

Whether MaxMind’s data is up-to-date/accurate is another matter…

16

Dave you’re right.
Wrong meaning from my side

Please schedule removal of your server as early as possible. We will stop putting your server in the pool two weeks before the scheduled removal date and stop monitoring and remove the server from the system on the removal date.

17

Do you have traffic statistics?

18

My router doesn’t have a notrack parameter. It does have traffic statistics, but they’re not enabled. That’s not an issue for now, as the cause of the problem became clear thanks to your answers.

The router can’t handle the increased number of connections caused by all the NTP clients in the area trying to connect to it. The issue is worse than I initially thought: not only does the score drop, but the entire internet connection for my office’s local network hangs.

Unfortunately, I won’t be keeping this NTP server active, as I can’t risk having my other servers, connected to the same network, become inaccessible. I’m very sorry for this situation.

19

Sorry to hear that the conditions don’t allow your server to join the pool, but fully understand.

When a country zone doesn’t have any servers in it, clients are automatically served from the encompassing continent zone. The continent zones are populated by servers allocated to any country zones in that continent zone that have a netspeed setting of 10Mbit/s or above.

I recently spun up a test server in Johannesburg, South Africa, and found that even with a netspeed setting of 12 Mbit/s (i.e., not only serving the ZA zone, but also the Africa zone), the resulting actual load was in the low triple-digit kbit/s range (might slightly increase going forward as more long-term clients associate with it).

So one thought that came to my mind, but which depends on your interest and enthusiasm to follow this through even with uncertain outcome, whether it is technically feasible/sensible, and whether you find an admin to effect that change.

But one option could be to drop your server from the country zone, where, being the only one, it will get overwhelmed by all the clients from that zone hitting it. And only leave it in the Africa continent zone, where the overall load is effectively shared by many more servers.

That way, your server could contribute to the pool after all, where “every server counts”. And you would even very much serve clients from your “country” zone, if that were a driver for you, as without any server in the country zone, they are currently being served by the Africa zone anyhow, which your server would then be a part of.

But again, don’t know the internals of the pool infrastructure as to whether that would feasible from a technical point of view, and work the way described above. It is possible for admins to move servers from one country zone to another, or assign a server to multiple country and continent zones simultaneously. Not sure whether having a server in a continent zone, but not in any country zone, is possible, e.g., from the “netspeed” setting point of view, and how the mechanisms responsible for assigning load to servers based on that setting work.

If interested, and if none of the admins sees this discussion, you could reach out to them via the e-mail address on the pool web pages, or the comment field offered when adding a server to the pool.